Analyzing Fidenza Artblocks

Important: This notebook uses Plotly mostly to visualize data. Plotly charts won't be rendered in Githubs notebook viewer. So please use nbviewer instead to view the notebook, or your local installation after cloning the repo.

FIXME Link

import plotly.io as pio
pio.renderers.default = "notebook_connected"

Get the Fidenzas

import datetime
from opensea import ARTBLOCK_CONTRACT, retrieve_assets
from IPython.display import display  # FIXME put elsewhere

FIDENZA_IDS = list(range(78000000, 78000998 + 1))
assets = retrieve_assets(token_ids=FIDENZA_IDS, contract=ARTBLOCK_CONTRACT)
print(datetime.datetime.now())

.................................................. -- All 999 assets retrieved.
2021-10-11 18:11:19.173898

Build a dataframe

import numpy as np
import pandas as pd
from helpers import turn_assets_into_df
df, traits = turn_assets_into_df(assets)

assert set(df.LPsymbol.unique()) == set(["ETH", "WETH", np.nan]), """
This sheet does not do currency conversion at the moment and therefore 
assumes all prices are in (W)ETH. But there are more symbols in the input 
data which would lead to apples being compared to ranges below. Aborting.
"""

How many pieces have a last price on OpenSea?

print(f"{df[~df.Lastprice.isna()].shape[0] / df.shape[0]:.0%}")

61%

How are prices evolving over time?

Note that this only takes into account the last sale of each piece.

import plotly.express as px
import plotly.graph_objects as go
import matplotlib.pyplot as plt

plt.rcParams["figure.figsize"] = (20, 8)
pd.plotting.register_matplotlib_converters()

fig = px.scatter(df, x="LPdate", y="Lastprice", hover_data=["Name", "Colors", "Probscore"])
fig.show()

Price levels today

fig = px.histogram(df, x="Lastprice")
fig.show()

How frequent are the different traits?

for trait in traits:
    display(
        pd.DataFrame(df[trait].value_counts(normalize=False, sort=True, ascending=True)).transpose()
    )

	Relaxed	Anything Goes	No Overlap
Collision Check	17	30	952

	Dark Lifestyle	Party Time	White on Cream	Luxe-Derived	Cool	Rose	Black	AM	White Mono	Baked	Politique	Rad	Golf Socks	Luxe
Colors	6	9	10	11	12	22	26	30	37	47	64	86	111	528

	Low	High AF	Medium	High
Density	49	68	277	605

	No	Yes
Have Margin	427	572

	Yes	No
Outlined	100	899

	Small	Jumbo XL	Medium	Micro-Uniform	Large	Uniform	Jumbo
Scale	14	30	35	38	181	189	512

	Sharp	Curved
Shape Angles	34	965

	Yes	No
Soft Shapes	150	849

	Yes	No
Spiral	35	964

	Yes	No
Super Blocks	241	758

	None	High	Low	Med
Turbulence	149	203	220	427

How much do people pay for the different traits?

Note that the charts are interactive. You can zoom, hover, pan, etc. Zooming is especially important due to the outliers in the data so you can zoom in on the interesting parts.

for trait in traits:

    fig = go.Figure()
    for traitvariant in (
        df.groupby(trait).median().sort_values("Lastprice", ascending=False).index
    ):
        fig.add_trace(
            go.Box(
                y=df[df[trait] == traitvariant].Lastprice.values,
                name=traitvariant,
                boxpoints="all",
                jitter=0.2,
                whiskerwidth=0.2,
                marker_size=2,
                line_width=1,
            )
        )
    fig.update_layout(title=trait)
    fig.show()

FIXME observations

Crosstabs

_ = pd.crosstab(df.Colors, df["Collision Check"]).plot(kind="bar", stacked=True)

Probscore as a simple unified metric to predict prices

The Probscore metric is simply the product of the probabilities of the different traits of a piece. There is currently not much value in that metric. While only the lowest probabilities are able to command the very high outlier prices, overall correlation between Probscore and Lastprice remains low:

fig = px.scatter(df, x="Probscore", y="Lastprice")
fig.show()
df[["Probscore", "Lastprice"]].corr()

	Probscore	Lastprice
Probscore	1.000000	-0.032479
Lastprice	-0.032479	1.000000

import plotly
plotly.offline.init_notebook_mode()